Machine learning-based compressive strength modelling of concrete incorporating waste marble powder

Abstract

In recent years, the volume of waste marble powder (WMP) from ornamental stone factories has increased rapidly, causing environmental concerns of soil, water and air pollution. While some studies have explored the benefits of incorporating WMP in concrete mixtures, the lack of pertinent data and a comprehensive understanding of how WMP influences the engineering properties of concrete has hindered the large-scale applications of WMP in the concrete industry. Therefore, this study examines the capability of machine learning (ML) to model the compressive strength (CS) of concrete incorporating WMP since it is the most specified property of concrete. WMP can improve the performance of concrete mainly owing to a physical microfiller effect and providing preferential sites for the nucleation and growth of early-age cement hydration products. However, considering that the pozzolanic reactivity of WMP is insignificant as demonstrated by phase composition and thermogravimetric results, its incorporation rate in concrete mixtures should be limited to a specific dosage, which can be optimized using ML techniques. ML modelling results showed that the CS of WMP concrete could be predicted with high accuracy (R2 > 0.97) using extreme gradient boosting (XGB) and artificial neural networks (ANN) informational models, with ANN being more sensitive to the content of WMP. The feature importance results, with multicollinearity consideration, showed that the role of WMP in strength development could be mainly limited to 10–20% due to its inert nature.